ReLooper: Refactoring for Loop Parallelism

In the multicore era, sequential programs need to be refactored for parallelism. The next version of Java provides ParallelArray, an array data structure that supports parallel operations over the array elements. For example, one can apply a procedure to each element, or reduce all elements to a new element in parallel. Refactoring an array to a ParallelArray requires (i) analyzing whether the loop iterations are safe for parallel execution, and (ii) replacing loops with the equivalent parallel operations. When done manually, these tasks are non-trivial and time-consuming. We present ReLooper, an Eclipse-based refactoring tool, that performs these tasks automatically. Experience with refactoring real programs shows that ReLooper is useful: it reduces the burden of analyzing and rewriting parallel loops, and it is fast enough to be used interactively.DOE grant ER25752Microsoft and Intel through the UPCRC Illinoisunpublishednot peer reviewe